I'll give an example I came across recently. They were trying to examine whether videogames could make someone more violent. So they got people to give two written samples before and after playing various types of computer games and analysed the language of them to see if there were differences. They did find a tendency for violent videogames to cause people to use more forceful or violent language (hard in itself to define, but whatever). The thing is, does this actually tell us anything about whether people are more violent after playing violent videogames? If I use more forceful language am I more likely to go out and hit someone over the head with a lump of wood or something? It's not like they can go into someone's head and see the likelihood of violent behaviour increasing so they needed to use a proxy, but I mean what can you use as a proxy for that? It's all very messy.